Contrastive representations in non-segmental phonology
Christopher Spahr
January 2016

This thesis develops and tests a unified model of word-level prosodic contrasts. Traditionally, word prosody has been analyzed within disparate models (such as autosegmental theory for tone, metrical theory for stress, and CV, X-slot, or moraic theory for length), meaning that it has not been possible to make clear predictions about how many different prosodic features can be employed in a single language. I present a minimal architecture for word prosodic representations based on a single set of formal elements. A tier of segmental root nodes, or X-slots, bears the binary contrastive features that divide the segmental inventory and represents quantity contrasts through two-to-one linking, while a tier of prosodic root nodes, or “pi-nodes”, bears the binary features dividing the autosegmental inventory. Features on pi-nodes are used in tone languages with more than one tonal autosegment, but in privative tone languages, the pi-node itself reflects the phonetic realization of the marked member of the tone opposition. The same featureless pi-node is used as an autosegmental marker of accented positions in lexical stress systems, where its language-specific phonetic realization is that of stress: some combination of increased pitch, duration, and intensity. The predictive power of this model is that it restricts systems to a maximum of two independent word prosodic contrasts, since each requires its own tier of root nodes. The pi-tier can represent either tone or accent separately from length on the X-tier, but this leaves no means to represent a third contrast. In certain systems, surface stress may be represented covertly as length on the X-tier with tone represented on the pi-tier, but no mechanism is available to host a third contrast, since the X-tier is already used for stress.